WPS6686 Policy Research Working Paper 6686 The Organization of Political Parties and the Politics of Bureaucratic Reform Cesi Cruz Philip Keefer The World Bank Development Research Group Macroeconomics and Growth Team November 2013 Policy Research Working Paper 6686 Abstract Bureaucratic reform is a priority of donor organizations, collectively to limit leader shirking. This is particularly including the World Bank, but is notoriously difficult the case with programmatic parties. Such politicians have to implement. In many countries, politicians have little stronger incentives to pursue public policies that require interest in the basic financial and personnel management a well-functioning public administration. Novel evidence systems that are essential to political oversight of offers robust support for this argument. From a sample bureaucratic performance. To explain this, this paper of 439 World Bank public sector reform loans in 109 presents a new perspective on the political economy of countries, the paper finds that public sector reforms are bureaucracy. Politicians in some countries belong to more likely to succeed in countries with programmatic parties that are organized to allow party members to act political parties. This paper is a product of the Macroeconomics and Growth Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at pkeefer@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team The Organization of Political Parties and the Politics of Bureaucratic Reform Cesi Cruz Philip Keefer* JEL codes: H11, H41, H83, 01, 019 Key words: political economy, political parties, public administration, public sector reform, economic development Disclaimer: The opinions and findings here are those of the author and do not represent the views of the World Bank or its directors. *Cruz: University of California, San Diego; cesi.cruz@gmail.com. Keefer: The World Bank; pkeefer@worldbank.org. The Organization of Political Parties and the Politics of Public Sector Reform Public sector performance is a central focus of international development assistance. About one-sixth of World Bank lending and advisory resources are dedicated to public sector reform, in areas ranging from civil service reform and budget preparation and planning to public sector financial management (World Bank 2008). These reforms have proven particularly difficult to implement, in part because politicians in many countries are indifferent to the performance of the public administration. They neither insist on public sector financial management systems that allow them to track agency spending, nor care that personnel management practices give them scant control over the quality of staff in the public administration. Instead, they allow lax financial controls and significant leakage, and patronage-based personnel systems and weak capacity to persist. The analysis here develops an original explanation for this phenomenon. Public sector reform offers collective benefits to politicians; individually, they have incentives to free-ride on reforms, subverting them for private political gain at the expense of politicians collectively. Where politicians are bound together by programmatic parties, they are more likely to have developed organizational arrangements that limit free-riding and promote collective action. Such politicians are also more likely to prefer broad public policies, which are particularly sensitive to the quality of public administration, and are better able to act collectively to oblige party leaders to implement financial controls and meritocratic personnel systems in pursuit of public sector reforms. We find support for this argument in tests that use unique data, evaluations of the performance of World Bank loans to support public sector reform. These evaluations are 2 significantly higher in countries where politicians are organized into programmatic parties: the probability that a public sector reform project is rated as successful is more than 30 percentage points higher in a country with all programmatic parties than in a country with none. Put another way, politicians in non-programmatic parties are significantly less likely to support formal public sector management systems that limit shirking by agencies and strengthen agency incentives and capacity to undertake the formal goals that politicians set for them. These tests provide the first systematic evidence that parties and party organization influence political incentives to embrace reforms to enhance the performance and oversight of the public administration. This conclusion contrasts with the bulk of the literature on the political economy of bureaucracy. Much of the US-focused research takes for granted that politicians care intensely about shirking in the public administration. Other research concludes that bureaucracies shape political incentives and mobilization strategies, rather than the reverse (Shefter 1994 and Chandra 2004). Tilly (1990) and others argue that politicians only permit strong bureaucracies when dramatic threats, particularly war, oblige them to do so, while our evidence points to political incentives to reform under more benign circumstances. The next section details the argument more thoroughly. One contribution of the research is to use new data on public administration reform projects from World Bank lending data to examine the politics of public sector reform; we describe these data in the following section. The remainder of the paper reports numerous tests using these data, supporting the conclusion that the organization of politicians into more programmatic parties increases political incentives to pursue public sector reform. The paper concludes by 3 highlighting some implications of these findings for research on the politics of public sector reform. Public administration and programmatic political parties At one level, the argument that programmatic parties favor the success of public sector reform, while new, is also intuitive. It might seem obvious, for example, that the policies preferred by programmatic parties place larger demands on the public sector; hence, their electoral fortunes are more likely to depend on the success of public sector reforms. This straightforward assertion begs key questions, however. What is it about programmatic parties that allows individual politicians to credibly commit to the pursuit of the party program? As the discussion below emphasizes, programmatic parties must be organized to resist politician incentives to free-ride on the party program. In addition, public sector reform generally gives party politicians outside the executive branch more opportunity to limit the discretion of their colleagues inside the executive branch. What is it about programmatic parties, but not other parties, that obliges executive branch officials to tolerate limits on their discretion? Individual politicians in programmatic parties have incentives to free-ride on public sector reform – for example, to ask the bureaucracy to allocate contracts and jobs to their personal constituencies, contravening formal public sector rules and at the expense of the party as a whole. What distinguishes programmatic parties from other parties in their ability to curb such free-riding? A final puzzle is that charismatic leaders often advocate broad public policies that could be better implemented if public sector reforms were undertaken. Nevertheless, these politicians typically resist public sector reform. The following sections address these questions by explaining in greater detail why the presence of programmatic parties increases political incentives to support public sector 4 reform. The first makes the case that public sector management reforms do play a larger role in the provision of government services to large groups of citizens, and therefore to politicians who mobilize political support with appeals to large groups of citizens. These reforms are less important, and even disadvantageous, to politicians with stronger incentives to pursue policies that target benefits to narrow groups of citizens. Second, politicians who can make credible commitments to large groups of citizens are more likely to pursue policies that benefit large groups of citizens. But, third, politicians organized into programmatic parties are more likely to be able to curb free-riding that would otherwise undermine both the credibility of these commitments and the ability of party members to press the executive branch to accept increased oversight over and limits to its discretion. Public sector management reforms matter more to politicians who seek to provide benefits to large groups of citizens Public sector management reforms generally focus on improving the formal systems of public administration along at least one of two dimensions. One dimension is to increase public sector capacity, for example by paying the top people relatively more (pay decompression), or by instituting stricter recruitment procedures that match job requirements to agency needs and hiring decisions to job requirements. The other is to permit greater external oversight of agencies or, as Moe (1987) and Cox and McCubbins (1994) argue, of the politicians who head the executive branch agencies. Opaque and lax public sector financial management systems, for example, might be strengthened to prevent individual officials and entire agencies from diverting public resources towards the pursuit of their own goals. Financial management systems allow all politicians, inside and outside of the executive branch, to observe the possible diversion of 5 financial resources away from their priorities. 1 By improving management information systems, politicians have access to more systematic data on the inputs and outcomes of public sector activity; integrating these data into budget and personnel decisions allows them to link the incentives of officials to the pursuit of approved agency goals. However, politicians do not always benefit from public sector reforms that control shirking by agencies and officials; systematically link budgets and promotions to the achievement of statutory goals; or introduce rigorous, objective criteria into the public sector hiring process. The benefits of such reforms are greatest when it is costly for politicians to personally verify that the public administration is implementing their policy objectives. These costs are, in turn, highest for politicians who are interested in delivering benefits to broad groups of citizens and lowest for clientelist politicians interested in transfers targeted at narrowly defined constituencies. With respect to external oversight, politicians with an interest in delivering benefits to many citizens confront greater monitoring challenges than those focused on narrower groups and individuals. No individual politician in the former group is in regular contact with more than a fraction of total beneficiaries and cannot, in the course of regular business, assess whether broadly targeted policies are being appropriately implemented. The latter group of politicians is focused on policies that benefit narrow groups of citizens with whom they are already in regular contact – they can easily track whether deserving groups have 1 On May 25, 2012, the director of the United States National Weather Service, an agency in the Department of Commerce, resigned after an audit found that the NWS had diverted approximately $30 million per year to 122 weather offices around the country without notifying Congress. Senator Olympia Snowe called the report “deeply troubling.” (The Washington Post, “Weather Service director retires after critical investigation of the agency,” p. A-2). 6 received promised benefits. They may even prefer to implement such policies themselves, as with constituency development funds (Keefer and Khemani 2009), in order to ensure that they, rather than the broader government, receive credit for delivering the benefits. A well- functioning public administration reduces the former group’s cost of monitoring more than the latter group’s. Policies with broadly-targeted benefits are also more demanding with respect to the information that politicians need to monitor official performance. To know whether the public administration is effectively implementing education programs, for example, politicians need information about student learning outcomes and the inputs to student learning that government education programs provide, as well as the capacity to distinguish the effects on student learning of unobservable family and student effort from the effects of public employee effort. These informational requirements are difficult for politicians to fulfill in the absence of formal administrative systems explicitly created for this task. Similarly, the provision of broad public policies requires the joint action of large numbers of public officials. Assessing the contribution of different officials to the desired outcomes is difficult for politicians, so they leave it to the public administration itself to make these judgments. In contrast, politicians focused on narrowly targeted transfers can easily observe whether their clients have received intended government benefits. Since only a few public officials are responsible for these transfers, politicians are less dependent on the administrative apparatus to reward the public officials who carry out their wishes; they can provide such rewards themselves, by recommending promotions, favorable postings or bonuses. 7 Finally, high administrative capacity for policy implementation matters less, in general, for narrowly- than for broadly-targeted policies. Politicians who care about the quality of health care and education benefit from experts who can identify the best modalities for delivering high quality and the population groups that are most likely to need subsidies. Politicians who seek to provide unemployment benefits that are generous enough to protect vulnerable populations, but not too generous so as to dissuade individuals from looking for new jobs, similarly benefit from experts who can analyze this tradeoff for them. In contrast, politicians whose main concern is channeling benefits to political supporters require no assistance either in identifying beneficiaries or in designing benefits. 2 In Pakistan, for example, there is almost no political interest in the broad economic effects of taxation, since most political attention is focused on creating ad hoc exemptions to tax rules. As a consequence, neither the Ministry of Finance nor the Federal Bureau of Revenue of Pakistan possesses the analytical capacity to estimate tax incidence and the effects of tax law changes on growth and, even, on revenues. 3 Not only is it the case that clientelist politicians benefit less from public sector reform. They may also believe that a well-functioning public administration will thwart their efforts to channel benefits to narrow constituencies. This is the logic underlying Shefter (1994), who argues that when new democracies inherit a well-functioning public administration, clientelist political strategies are stymied. Civil service reforms intended to introduce Weberian, meritocratic principles into civil service recruitment and promotion 2 Consistent with this line of argument, Johnson and Libecap (1994) argue that Congress professionalized the civil service after the Civil War and its aftermath had increased the scope and complexity of government activity. 3 Author interviews, Federal Revenue Bureau of Pakistan, February 2013. 8 standards, for example, militate against the use of the civil service as a vehicle for patronage (see Ishii, Rohitarachoon and Hossain no date, for example). Similarly, transparent and rigorous financial management and procurement systems make it more difficult to use government contracts to reward supporters. In contrast, politicians who are interested in delivering high quality services to the largest number of citizens are more likely to care about procurement systems that drive down the costs of service provision. 4 Policy discussions regarding public sector reforms emphasize their importance for the delivery of broadly-targeted government policies. A note outlining the World Bank strategy for public sector management and governance reform states that “Improving public sector management is ultimately about providing firms and households with better services. . .”, such as health, education, infrastructure, and regulations to improve, for example, food and transportation safety (World Bank, no date, p. 1). Politicians who seek to improve the quality of broadly-targeted services to voters should therefore also support public administration reform. Politicians able to make broadly credible commitments are more likely to support policies that provide benefits to large groups of citizens Countries differ significantly in the degree to which their politicians aim to mobilize support around narrowly clientelist or broader, programmatic appeals. One explanation for this difference is variation among politicians in the degree to which they can make credible pre-electoral appeals to large groups of citizens. 4 Huber and McCarty (2004) argue that politicians’ uncertainty about competence reduces their ability to monitor the public sector. Civil service reforms, such as meritocratic hiring and promotion standards, reduce this ambiguity. Their model assumes that politicians are concerned about the quality of public policy implementation; when this concern is absent, so is the incentive of politicians to reduce uncertainty about competence. 9 A standard result in a large literature is that political incentives to provide public goods depend on whether they need the support of large constituencies to remain in office (e.g., Persson and Tabellini 2000). The larger the constituency, the more politically efficient it is to deliver benefits in the form of public goods rather than private transfers. Money spent on a public health program to control infectious disease in a large group can raise average citizen welfare by much more than if politicians use the same resources to deliver small cash transfers to each citizen. If the group is small, however, the same resources translate into large cash transfers and, from the point of view of the small group, the benefits of private transfers may exceed those offered by the public health program. A substantial body of research, including Persson and Tabellini (2000), links the size of politicians’ constituencies to political and electoral institutions. For example, McCubbins and Rosenbluth (1995) emphasize that single-member districts increase the importance of politicians’ personal characteristics in voters’ electoral choices, rather than the characteristics of the group of politicians with which they are affiliated. However, cross-country differences in electoral rules are not robust predictors of the policy choices that countries make. For example, they do not robustly distinguish the policy choices of younger and older democracies in Keefer (2007), despite the fact that the former are disproportionately presidential. Keefer and Vlaicu (2008) point instead to the ability of politicians to make broadly credible political promises to explain why politicians in some countries prefer policies that favor only narrow clienteles. “Clientelist” politicians, those who can make credible pre- electoral commitments only to narrow constituencies, prefer narrowly targeted policies to public good provision. Politicians capable of making pre-electoral commitments to broad 10 groups of citizens, in contrast, care correspondingly more about policies that deliver benefits widely. This group should also support reforms that improve the public administration. Keefer (2011, forthcoming) argues that the inability of politicians to act collectively also undermines their ability to make credible promises to broad groups of voters. Collective action is essential because individual politicians have neither the ability nor the incentive to deliver on broad public policy promises. They lack the ability because multiple politicians – in both democracies and non-democracies – must cooperate in the approval and implementation of policies aimed at delivering benefits to large groups of citizens. They lack the incentive because, as in the canonical collective action setting, they bear the full cost, but receive only a fraction of the benefits, of their individual actions that promote party success. For example, they bear the full cost of depriving their narrow constituencies of policy benefits in order to further the policy goals advanced by the party as a whole. Programmatic parties are more likely to exhibit the organizational characteristics that allow politicians to make credible commitments to large groups of citizens Absent a vehicle for coordinating politician actions and limiting the incentives of individual politicians to free-ride on the efforts of others, public policies aimed at delivering benefits to large groups – including the administrative apparatus necessary to implement these policies – are less likely to emerge. Political parties can solve the coordination problem, particularly if they exhibit two organizational characteristics (Keefer 2011, forthcoming). First, they authorize leaders to monitor and punish members who shirk in the pursuit of the party’s collective interests. Second, though, because party leaders do not necessarily fully internalize party interests, the parties are also organized to allow members to act collectively to control shirking by leaders. 11 Programmatic parties are more likely to exhibit organizational arrangements that curb free-riding and allow members to impose greater constraints on leaders. Public sector reform is correspondingly more likely to succeed in the presence of these parties. It is well- known that programmatic parties allow citizens to hold politicians collectively accountable for their actions, encouraging politicians to care less about the particularistic benefits they deliver to voters and more about broadly distributed benefits (Aldrich 1995). However, parties can only perform this function if they are internally organized to prevent free-riding by members and shirking by leaders in their pursuit of the party program. Since the party program is a collective good, individual members have an incentive to shirk on it; since the private interests of leaders are rarely perfectly aligned with the interests of the party, they also have incentives to deviate from the program. Programmatic parties succeed when they adopt organizational arrangements to control free-riding and shirking (Keefer 2011, forthcoming). The organizational attributes that allow programmatic parties to overcome free-rider problems in the defense of the party program also allow them to solve free-rider problems inherent in public sector reform itself. Reforms of the public administration are a public good from the point of view of both society at large, but also politicians themselves. Rule- based financial and personnel management systems offer collective benefits to all politicians. Individually, however, politicians have an incentive to free-ride on the creation and maintenance of such systems, so public sector reforms are more likely to succeed when politicians are organized to limit free-riding. They also have an incentive to free-ride on costly actions to compel party leaders to accept administrative reforms. 12 Moe (1987) and Cox and McCubbins (1994) both emphasize that, absent the ability of non-executive politicians to act collectively and to punish “defectors”, executives can use selective punishments and inducements to persuade individual legislators to give up their proposals to reform the public administration. They make these arguments in the context of presidential political systems. However, legislative members of governing parties in parliamentary systems confront a similar challenge if they seek administrative reforms that enhance their ability to supervise cabinet officials – who are also, generally, party leaders. Programmatic parties, therefore, favor the success of public sector reform because they exhibit the organizational characteristics needed to overcome free-riding by party members, and shirking by party leaders, to ensure the credibility of the program. These same organizational traits overcome free-riding in the implementation of public sector reform itself and allow party members to impose such reforms on party leaders in the executive branch. While programmatic parties favor the success of public sector reforms, is it the case that non-programmatic parties have organizational characteristics that disfavor them? In fact, a large class of non-programmatic parties has unambiguously weak incentives to pursue public sector reform. One type of party that is heavily represented among non- programmatic parties consists of loose associations of patron-politicians. These politicians can individually make credible commitments to their personal client network. However, their parties lack the organizational arrangements necessary to punish free-riding because such politicians lose little by switching parties. Party leaders cannot easily prevent them from pursuing policy objectives that are contrary to those advocated by the party as a whole and the candidates themselves have little incentive to act collectively to oversee party leaders. 13 Lacking the organizational ability to overcome free-riding by party members or shirking by leaders, they have no capacity to credibly commit to broadly-targeted public goods. The mobilization strategy of these therefore rests on individual patrons’ ability to attract their clients’ support with promises of narrowly targeted benefits (see, e.g., Bratton and van de Walle 1997, Kitschelt and Wilkinson 2006 and Keefer and Vlaicu 2008). The incentive and capacity of these politicians to build an effect administrative apparatus is weak, since they can personally oversee whether private government benefits flow to their constituents. Public administration reform is therefore less likely to succeed in countries governed by parties in which politicians can only make credible commitments to their private networks. Machine parties are also likely to prefer clientelist policies, targeted at narrow groups, but with the important difference that the provision of private goods is targeted to party supporters and mediated by the party. Machine parties have more ambiguous incentives with respect to the pursuit of public sector reform. On the one hand, machine parties exhibit substantially more organizational capacity to overcome free rider problems than loose associations of politicians, each with personal constituencies. The party machine works because party leaders can impose sanctions to limit shirking by party members who fail to use party resources to mobilize electoral support. 5 As a consequence, following the logic of Aldrich (1995), voters are better able to hold party politicians collectively 5 The ability to impose such sanctions is not a trivial organizational hurdle. Monitoring by party leaders is difficult (see, e.g., Kitschelt 2000 on the Austrian ÖVP). To accomplish it, machine parties exhibit sophisticated organization that carefully tracks members’ contributions to the party’s electoral success, offering privileged access to private transfers from the party or government to members who contribute more (Kitschelt 2000; Stokes 2003; Calvo and Murillo 2013, forthcoming). 14 accountable for government failures than they are in the case of loosely organized coalitions of patron-politicians. Generally, however, these parties do not use their organizational capacity to maintain a credible party program. As Kitschelt (2000) observes, machine parties typically do not invest in the organizational infrastructure of interest aggregation. Keefer (forthcoming) points to two reasons for this. On the one hand, the party’s own comparative advantage in providing targeted benefits to voters is undermined to the extent that the public sector is effectively organized to provide broad-based public goods. On the other hand, recruitment into machine parties tends not to hinge on the programmatic preferences of potential members. When the policy preferences of machine party members are heterogeneous, though, organizational features of the party that allow them to act collectively are unlikely to be sufficient to establish a credible programmatic line for the party – party members are too heterogeneous with respect to policy preferences to agree on a program. Machine parties are therefore less likely than programmatic parties to pursue public policies that benefit from public sector reform; and they are less likely to solve the collective action problems needed to allow members to avoid free-riding in the pursuit of public sector reform itself. Parties may also organize around the leadership of charismatic individuals, and rely on the leaders’ charisma to mobilize electoral support. These leaders may advocate a broad, programmatic agenda and, therefore, stand to gain from a well-performing public sector. However, they have less to gain from either well-organized bureaucracies or well-organized ruling parties: collective action by either poses a potential challenge to their authority. Moreover, direct involvement in the implementation of the policies they advance enhances 15 the charisma of these leaders. By delegating implementation to a well-organized public administration, they dilute their charismatic appeal. Hugo Chávez, for example, advanced significant social programs that largely benefited the poor and were not strongly conditioned on overt support for the ruling party (Hawkins 2010). His party, though, like those of other charismatic leaders, was organized around the personality of the leader, with little institutional basis of its own (Zúquete 2008, Roberts 2003). The internal organization of such parties does not impose costs on leaders if they fail to pursue redistributive policies in the future. This is because voter support for the party rests not on the credibility of the party’s commitment to redistributive policies, but on their faith in the charismatic leader’s commitment to those policies. Indeed, it is precisely because Chávez’ personal charisma was sufficient to persuade voters of his credibility that he could refrain from introducing key organizational arrangements into the Partido Socialista Unido de Venezuela that would have allowed members to act collectively to restrain leader shirking. Knowing that it is not optimal for them to improve the public administration, charismatic leaders (like Chávez) therefore emphasize administratively easy transfers. Even when they seek to deliver transfers to the poor, consistent with their promises, they tolerate loose targeting (they are willing to miss some of the poor and send resources to the less poor), since more direct targeting would require a stronger administrative apparatus. Similarly, they avoid the administratively difficult task of improving service quality (or, as in the case of Chávez, improve service quality by importing doctors from Cuba, essentially free- riding on the administrative capacity of allied countries). 16 The analysis of charismatic leaders applies more generally to country executives, whether elected or unelected. They confront challenges in overseeing the bureaucracy over which they preside and must decide whether to meet these challenges by adopting transparent, rule-bound financial and personnel management systems. Charismatic leaders are likely to resist public sector reforms because they prefer not to allow the organization of their party, which would otherwise serve as a counterweight to a well-organized public sector. Executives who are not charismatic, either elected or unelected, face greater obstacles to mobilizing political support. If they wish to mobilize support with credible promises to provide broadly-targeted policies, rather than narrowly targeted, more clientelist transfers, then they benefit from a ruling party organized to convince citizens of the credibility of those commitments and from a public administration organized to implement those policies. 6 In contrast, Joaquín Balaguer, the president of the Dominican Republic for many years, did not represent a programmatic party and, therefore, personally directed the spending of as much as 50 percent of the national budget (Keefer and Vlaicu 2008). Testable hypothesis and data: Programmatic parties and World Bank public sector reform projects The foregoing argues that politicians more interested in policies targeting broad groups of citizens are also more likely to support well-functioning public administrations; that groups of politicians who are able to overcome individual incentives to free-ride have stronger incentives to provide broadly targeted policies and support institutions that 6 Gehlbach and Keefer (2011) show that organizational arrangements that ensure the free flow of information about ruler treatment of individual party members are sufficient to make ruler commitments not to expropriate party members credible. In an extension, they also show that the same logic applies to the public administration. Leaders would like to provide selective benefits to public officials who exert effort in the pursuit of goals established by the leader, but the promise of these benefits is only credible when the officials area aware of how the leader has treated other officials. 17 strengthen the public administration; and that clientelist parties organized around individual patron-politicians are least likely to be able to exhibit organizational characteristics that support collective action by politicians. Using data from World Bank evaluations of public sector reform projects, we test a reduced-form prediction that emerges from this argument: politicians organized into programmatic parties have stronger incentives to strengthen the public administration. The analysis here exploits a new source of data that precisely characterizes the public sector reforms undertaken by countries. These are evaluations of World Bank public sector reform loans, produced by the Independent Evaluation Group, an independent unit of the Bank that reports to the Bank Board of Executive Directors and not to Bank management. IEG evaluations are commissioned by the Board or decided upon unilaterally by the IEG. Bank management cannot require any changes in the wording or conclusion of the IEG document or its underlying ratings. We use IEG evaluations of 511 public sector loans that began in 1987 or later and ended in 2006 or earlier. World Bank loans often have multiple, cross-sectoral objectives; the IEG data allow us to focus on those loans that were most related to the types of public sector reforms that limit shirking by public officials and agencies, improve incentives for officials to pursue statutory goals, and improve the quality of officials in the public administration. Each of these maps directly into the earlier discussion (see the Data Appendix for a full explanation of the data). In particular, of the 511 loans, 439 were focused on public sector administration, public sector financial management, or tax administration (as opposed to, for example, judicial reform or decentralization). 18 Two IEG ratings are of particular interest. The most relevant is the IEG rating of the contribution the project made to institutional development. Since public sector reforms are precisely aimed at institutional development, this is the most apt measure of success. The other is the IEG rating of overall loan success. These are ordinal rankings. We collapse each into two categories. The institutional development indicator variable takes the value of one when the institutional development impact of the project is “substantial” or higher and zero otherwise. The overall outcome indicator variable takes the value of one when the outcome of the project is “satisfactory” or higher and zero otherwise. 7 Our argument is that the presence of programmatic parties increases political incentives to pursue public sector reform. Other theories make the reverse prediction, however, that organized public sectors increase political incentives to make programmatic appeals to voters. Shefter (1994) and Chandra (2004), for example, conclude that parties turn to programmatic appeals in countries where bureaucracies are independent, well- functioning and resist the implementation of arbitrary, clientelist policies. Though they do not identify the specific features of bureaucratic organization that make them independent and well-functioning, it is likely that these are related to the bureaucratic institutions on which we concentrate, those that facilitate monitoring, link incentives to statutory agency goals, and improve capacity. Our data allow us to exclude reverse causation as an explanation for our results. They refer to public sector reforms and the effect of parties at the outset of those reforms, excluding the possibility that the reforms have shaped the parties. 7 The main results are robust to alternative thresholds for project success and to the use of ordered logit regressions using the full ratings scales. 19 Moreover, all of our countries exhibit weaker bureaucracies and all of our specifications control for initial bureaucratic quality. 8 Previous cross-country research on the causes and effects of bureaucratic quality (e.g., Keefer and Knack 1995, Rauch and Evans 2000) has relied on subjective and broad evaluations of bureaucratic quality, particularly measures of corruption and bureaucratic quality in the International Country Risk Guide. Keefer (2011) shows that these are, in fact, associated with programmatic parties. However, since the focus of these data is on the performance of the public sector, they provide no information on the characteristics of the bureaucratic systems that underlie performance: whether public sector financial management systems are capable of tracking spending, or whether personnel management systems are designed to attract and retain the best and brightest candidates for public sector work. With our data, we are able to focus on specific aspects of bureaucratic organization that politicians can directly influence, abstracting from historical legacies, the socio- economic environment and other unobservables that are not under the direct control of politicians. A measure of programmatic parties The main correlate of interest is the presence of programmatic parties in a country and is constructed from variables in the Database of Political Institutions (DPI, Beck et al. 2001). They identify whether each of the largest three government parties and the largest 8 Our interpretation is also consistent with historical and comparative evidence that even strong bureaucracies often do not constrain political strategies. For example, despite the strong bureaucratic legacy of British colonial rule in India and Kenya, clientelist appeals are pervasive. Van de Walle (2003) argues that in the 1990s, parties in African democracies switched to programmatic appeals not because strong bureaucracies pushed them away from clientelist strategies, but because politicians themselves found it difficult to manage clientelist handouts. 20 opposition party are right, left or center in their orientation or whether, on the contrary, their orientation is either not discernible in the sources employed or unrelated to economic policy. Our key variable is the share of these parties that is identified as having a right, left or center orientation (rlc_avg). Across all the loans in our sample, about half of all parties do not exhibit such an orientation. We use the value for rlc_avg that corresponds to the first year of the loan, which is unlikely to be influenced by the successful or failed implementation of the loan and makes it more difficult to attribute our findings to a pattern of causality running from bureaucratic quality to programmatic parties. Results are robust to other timing assumptions. 9 The coding of the DPI is based on descriptions of the party in yearly political almanacs. When these descriptions indicate that the party pursues a more liberal, right-wing economic agenda, enumerators classify it as “right”. When they indicate that the party pursues a more left-wing economic agenda – is more redistributionist, oriented towards greater regulation of the economy, or favors legislation that privileges workers over firm owners – enumerators classify it as “left”. They code the party as centrist when sources indicate that the party’s mobilization strategy relies on elements of both a left- and right- wing economic agenda. When, however, parties are not organized around an economic program, and instead around the promotion of an individual leader, the party is coded as having no economic program. The formal coding rules aim at detecting whether parties are programmatic or not, reasonably classifying as programmatic the African National Party in South Africa, the main 9 Such as averaging rlc_avg over the project period or using 3 and 5 year backward averages. 21 Chilean parties, the Mexican PAN and PRD, the Brazilian Worker’s Party, and APRA in Peru. Moreover, the DPI tends not to code as programmatic those parties that are not organized to support a programmatic agenda, even though party leaders pursue one. Hugo Chávez’ Partido Socialista Unido de Venezuela (PSUV), is not coded as programmatic in the DPI, because it is organized around the personality of the leader, with little institutional basis of its own. Other parties with a more uncertain programmatic content and that rely more on clientelism to mobilize support, such as the Peronists in Argentina or the National People’s Party in Ghana, are coded as programmatic. Again, however, this is consistent with the earlier discussion. In contrast to other clientelist parties, particularly in their regions, these parties have more sophisticated organizational arrangements to limit member free-riding. Compared to more loosely-organized clientelist parties, which are more reliant on the ability of individual politicians to mobilize votes, they also give members greater sway over leader selection – visible, not least, in the fact that these parties have experienced leadership transitions even when incumbents were still alive. The DPI data on programmatic parties are not, however, sufficiently nuanced to classify parties according to the underlying strength of their organizational arrangements to enforce the party program. All purely programmatic parties that exhibit a high level of mutual control by members and leaders of each other’s potential shirking are certainly classified as programmatic. Similarly, all loosely organized associations of clientelist politicians are classified as non-programmatic. The differences between these two allow for the effects of programmatic parties to be identified from the data despite the presence of a noisier group of countries, such as those with machine parties that are coded as being 22 programmatic despite the fact that the party organization to enforce the program is subordinated to maintenance of the party machine; or machine parties coded as non- programmatic, despite the fact that they are significantly better organized to curb free-riding within the party compared to loosely organized clientelist parties. To show that noise in the DPI coding injects a bias against our hypothesis, we examine the robustness of our results to plausible re-codings of the variables. 10 We also examine whether results are robust to dropping questionable cases altogether. Since the DPI coding of party programs are based on political almanacs and, therefore, likely to be noisy, but unbiased, re-codings based on more detailed and specific country knowledge should reduce noise and strengthen our results. In fact, the plausible re-codings that we undertook have this effect. Information from other sources supports the interpretation of the DPI indicator as a measure of a party’s ability to make credible programmatic appeals to citizens. The Duke political parties database has extensive information on parties in 80 democracies in 2009. One can compare these variables to the average DPI programmatic variable of sample countries with projects during period since 2000. The comparison points to a high correspondence between rlc_avg and the more detailed party characteristics identified by expert respondents in the Duke database. For example, if parties are programmatic, one would expect them to have more stable factional groups in the legislature – party-switching is more difficult for politicians when ideological labels are attached to party affiliation. The 10 We did not undertake a comprehensive re-examination of the DPI party database, but we did observe three countries where our own country knowledge pointed to plausibly more accurate codings, at least relative to other countries in their regions. However, our results strengthened when we recoded Mexican parties to be more programmatic than Argentine parties; Ghanaian to be more programmatic than other sub-Saharan African parties; and Golkar in Indonesia to be more programmatic than all Philippine parties. 23 correlation with factional instability variable in the Duke database (question A.7) with rlc_avg is -.42 for the 48 countries that are common to our database and the Duke database over this period. Programmatic parties should also rely less on charismatic leaders to mobilize voters. The Duke variable indicating whether parties are led by charismatic leaders (E1) is significantly negatively correlated with rlc_avg (-0.28). Parties can be programmatic on other margins, such as nationalism or religion. As Kitschelt (1993) observes, when parties have highly ideological members, the incentives of members to free-ride on the party program are lower. To the extent that nationalist and religious parties have more ideologically-motivated members, therefore, it is easier for them to curb free-riding, even without investing in organizational arrangements to prevent it. However, the public policies for which a well-performing public sector is useful may also be less salient to such parties. It is therefore ambiguous whether these parties would exhibit weaker incentives to support public administration reforms than the “economically” programmatic parties on which we focus. There are few such parties in our sample; in the main analysis below, they are included in the non-programmatic category. However, our results are robust to moving nationalist parties from the non-programmatic to the programmatic category. Control variables The analysis takes into account a number of country and loan characteristics that might exert a spurious influence on the relationship between World Bank project ratings and the presence of programmatic parties. All of the variables used here are summarized in Table 1. Poorer countries are less able to accumulate the human capital needed to run an efficient public sector and to implement public sector reforms successfully. Countries with 24 more poor people may also, simultaneously, be more susceptible to the allure of clientelist promises of individual transfers. All regressions therefore control for the log of per capita income from the World Development Indicators, averaged over the duration of the loan. 11 The size of a loan can also affect its ratings. The larger a loan, the greater are the resources that are potentially applied to public sector reform, which may improve ratings. Larger loans may exhibit greater sectoral diversity, making the ratings noisier measures of public sector reform success. Larger loans may also attract stricter and more accurate scrutiny by the IEG, however, making them less noisy. All regressions therefore control for the loan amount and the percentage of the loan amount that is allocated to the public sector. Estimations also include a control for the duration of the loan. Longer duration loans could imply a more conscientious effort by countries to pursue loan objectives (consistent with the frequently-voiced argument that institutional reform takes longer). However, within the World Bank, slow loan disbursements can also be a red flag, indicating difficulties with implementation. In either case, duration is a rigorous control variable, since it may capture variation in IEG ratings that is actually due to political decisions linked to the presence or absence of programmatic political parties. Larger countries may be more difficult to administer; on the other hand, there may be economies of scale in administering more populous countries (holding area constant). There is, in fact, some evidence that public sector reform is more likely to succeed in countries with more people. Therefore, all regressions control for population and land area, 11 The main results hold when we use project start year values instead of averages over the term of the loan. 25 from World Development Indicators, as well as for region dummies, using the UN standard regions. Countries’ initial level of bureaucratic quality may influence their ability to earn a successful rating from the IEG. Moreover, the specific characteristics of public sector reforms may depend on initial bureaucratic quality. Our main results are robust to controlling for the beginning-of-period evaluation of bureaucratic quality found in Political Risk Services’ International Country Risk Guide. Donor agencies argue that project outcomes depend on project preparation: the better is project preparation, the more likely the project is to succeed. IEG provides ratings of project quality at entry and results are robust to its inclusion. This is again a conservative test, however. The quality of loan preparation is related not only to Bank inputs, but also to country engagement with the lending process. Engagement, though, should be a function of political incentives. Quality at entry could therefore also capture variation in loan success ratings that is more properly attributed to the presence or absence of programmatic political parties. All of the foregoing variables appear in our main, base specifications. We also examine the robustness of our results to numerous other controls. A large literature argues that formal political and electoral institutions influence political incentives to pursue particularistic versus public goods. 12 However, the introduction of institutional controls may lead to the spurious rejection of the hypothesis, since many theories suggest that these institutions influence political incentives to build programmatic political parties. For 12 Nearly all of that literature assumes that political parties are able to credibly promise public goods, and are therefore programmatic. The effects of these institutions in settings where political parties are weak and politicians are not collectively organized are less clear. 26 example, Cox (1997), Strom (1990), and Cox and McCubbins (1993) all argue that electoral rules affect the cohesion and discipline of parties. Bowler, Farrell, and Katz (1999), Cox (1997) and Cox and McCubbins (1993) also emphasize that parliamentary systems increase incentives for party cohesion. Nevertheless, although the political system variable is, in particular, significantly related to our measure of programmatic parties, results are robust to controls for all these variables. Table 1: Summary Statistics Standard Count Mean Minimum Maximum Deviation IEG, Institutional 295 .41 .49 0 1 Development Impact Rating IEG, Overall Outcome Rating 301 .42 .49 0 1 Programmatic Parties 415 .52 .38 0 1 Log GDP per capita 418 1.27 1.66 .08 9.02 Log Population 418 16.59 1.47 12.95 20.94 Log Land Area 426 12.63 1.64 5.83 16.61 Funding Amount 439 99.26 198.33 .49 2525.25 Project Duration 439 3.78 2.53 1 12 IEG Quality at Entry 305 4.21 1.42 1 6 ICRG Bureaucratic Quality 354 2.51 1.13 0 6 Loan % Allocated to PSR 439 44.64 21.01 11 100 Newspaper circulation 321 34.55 42.18 .09 200.15 Gross secondary school 411 48.72 30.06 5.19 110.36 enrollment Natural Resource Exports (% 365 2.69 5.38 0 49.19 GDP) PR 401 .54 .48 0 1 Political System 423 .39 .74 0 2 Mean District Magnitude 393 15.19 34.19 .7 281.5 27 Democracy (ACLP) 423 .55 .48 0 1 Bureaucratic reform may be easier in countries with greater human capital, so we therefore check for robustness to controls for gross secondary school enrollment. Again, this control can lead to the spurious rejection of the hypothesis, since education is precisely the type of broad-reaching policy that politicians in programmatic parties should be more likely to support. Results, though, are robust to its inclusion in the base specification. Most discussions of public sector organization and conflicts of interest between executive and legislature emphasize information asymmetries. We use newspaper circulation in countries, taken from World Development Indicators, to account for this effect. Gehlbach and Keefer (2010) argue that leader willingness to allow for collective action (by a ruling party or by a bureaucracy) is attenuated when natural resource rents are high. Knack (2008) shows that natural resource exports weaken government incentives to design efficient tax systems. We therefore control for natural resources as a percent of exports, from WDI. Specification and results The dependent variables are dichotomous, so we use a logistic specification to estimate the effect of party organization: = 0 + 1 ∗ ( ) + X 2 + . Each observation i is a loan. We estimate only clustered standard errors, assuming that errors are distributed independently across countries, not across loans within countries. Xi is a vector of covariates; εi is an error term; and β0, β1, and β2, are (vectors of) parameters to be estimated. Table 2 reports the results of logistic regressions for the baseline and full models. The baseline control variables include region dummies, country characteristics (GDP per 28 capita, population, land area) and project characteristics (project duration, funding amount, loan percentage allocated to public sector reform), and the full model adds IEG quality at entry and initial bureaucratic quality to the base model. All tables report odds ratios, the ratio of the odds that a project earns a successful rating to the odds that it earns an unsuccessful rating, adjusted for the simultaneous effect of the other explanatory variables. 13 Odds ratios greater than one imply that increases in the independent variable raise the probability of success; less than one, reduce it. Table 2: Programmatic Parties and Project Ratings for Public Sector Reform Base Model Full Model Base Model Full Model Inst. Dev. Impact Rating Overall Outcome Rating ** ** Programmatic Parties 3.874 6.061 2.358* 2.616+ (0.001) (0.004) (0.020) (0.083) Log GDP per capita 1.165 1.133 1.079 1.108 (0.315) (0.385) (0.474) (0.424) Log Population 1.570+ 1.874+ 1.275 1.402 (0.066) (0.078) (0.131) (0.180) Log Land Area 0.832 0.768 0.848 0.814 (0.350) (0.286) (0.235) (0.300) Funding Amount 0.998+ 0.998* 0.998* 0.998* (0.092) (0.046) (0.042) (0.026) Project Duration 0.968 1.066 0.874+ 1.013 (0.618) (0.350) (0.051) (0.852) Loan % Allocated to 1.005 1.008 1.007 1.011 PSR (0.411) (0.313) (0.290) (0.232) IEG Quality at Entry 1.990*** 3.505*** (0.000) (0.000) ICRG Bureaucratic 0.659+ 0.598* 13 For all countries in the sample, the odds ratio (odds of success relative to the odds of failure) is .653 for institutional development impact and .797 for overall outcome: failure is more likely than success. The corresponding probability that a project will earn a successful rating is .395 for institutional development impact and .445 for overall outcome. 29 Quality (0.057) (0.016) Region Dummies Yes Yes Yes Yes Observations 277 235 282 240 Pseudo R2 0.169 0.302 0.079 0.306 Logistic regressions; odds ratios reported; p-values in parentheses + p < 0.10, * p < 0.05, ** p < .01, *** p < .001. GDP per capita and population are averaged over the project period; progression to secondary school is a country average; programmatic parties and ICRG bureaucratic quality are measured at the project start year Figure 1: The Partial Effects of Programmatic Parties on the Probability of Successful Institutional Development Ratings on Public Sector Reform Loans (from Table 2, Column 1) 30 Figure 2: The Partial Effects of Programmatic Parties on the Probability of Successful Public Sector Reform Loans (from Table 2, Column 3). The odds ratios for programmatic parties in the base model are 3.87 for the rating on institutional development impact and 2.36 for the rating on overall project outcome. This implies that a one unit change in the proportion of programmatic parties increases the odds that a project earns a successful institutional development impact by 3.9 times relative to the odds that it earns an unsuccessful rating; and the odds that it earns a successful overall outcome rating by 2.4 times relative to the odds that its rating is unsuccessful. When controlling for initial bureaucratic quality and the quality of the project design in the full model, the odds ratios are larger still: 6.06 for institutional development impact and 2.62 for overall project outcome. These effects are illustrated in Figures 1 and 2. Another way to see the magnitude of the party effect is to hold all other variables at their mean values and to calculate the effect of an increase in the proportion of programmatic political parties from zero to one on the probability that a project is rated a 31 success. Increasing the proportion of programmatic parties from zero to one raises the probability of institutional development impact success from .240 to .560 in the base model and from .197 to .591 in the full model. For overall project outcome ratings, the probability of earning a successful rating rises from .275 to .525 in the base model and from .188 to .449 in the full model. The only baseline control variable that is significantly associated with both IEG ratings is funding amount: larger loans are associated with a lower probability of loan success. This is not due to a dilution effect – that public sector reforms play a smaller role in larger loans – because we control for the percent of the loan that the Bank attributes to the PSR component. Of the two additional variables in the full model, the IEG rating for quality at entry is significant and positive for both the project outcome and institutional development impact ratings. On the other hand, initial bureaucratic quality is associated with a decrease in the probability of a project earning a successful outcome rating. This estimated effect is consistent with at least three possibilities: that reform is more difficult in more advanced bureaucracies; that the Bank may hold countries with lower quality bureaucracies to more lax standards; and that public officials in more institutionalized bureaucratic settings may have greater ability to resist reform. The results in Table 2 are not driven by extreme observations. If anything, the exclusion of outliers strengthens them. This is true excluding those observations that had the largest residuals after plotting actual versus predicted values; those that, according to Cook’s D, have the greatest overall influence on the regression results; and those with large dfBetas, which have the largest influence specifically on the programmatic party coefficient (results not reported). 32 Table 3 demonstrates robustness to the inclusion of a broad range of additional control variables. The main results hold even when controlling for gross secondary school enrollment, newspaper circulation and natural resource exports, and for the inclusion of political controls (whether the electoral system is proportional representation; whether the political system is presidential, semi-presidential or parliamentary; mean district magnitude; and democracy, as determined by ACLP). Other results (not reported) reject two additional alternative hypotheses. One is that countries with programmatic parties have different policy preferences than those without and that it is these preferences that drive the public sector reform results. We investigate this possibility by controlling for whether the main government party is right, left or center; our main variable of interest remains highly significant. Second, it might also be the case that party fractionalization and divided government both increase incentives to form programmatic parties and introduce conflicts of interest between different branches of government, increasing the demand for public sector reform. Our results are robust, however, to controls for party fractionalization and divided government (defined as whether the executive is from a different party than the majority party in the legislature). Table 3. Programmatic Parties and Project Ratings for Public Sector Reform-- Additional Controls Media, Political Media, Political Education, and Controls Education, and Controls Natural Natural Resources Resources Inst. Dev. Impact Rating Overall Outcome Rating Programmatic 6.629** 2.766* 2.815* 2.302+ Parties (0.002) (0.033) (0.011) (0.062) Newspaper 0.999 1.005 circulation 33 (0.925) (0.399) Gross secondary 0.991 0.983 school enrollment (0.607) (0.239) Natural Resource 1.034 0.931 Exports (% GDP) (0.377) (0.211) PR 0.893 1.403 (0.775) (0.409) Political System 1.372 1.159 (0.230) (0.479) Mean District 0.994 1.005 Magnitude (0.484) (0.454) Democracy (ACLP) 1.090 0.741 (0.868) (0.473) Baseline controls Yes Yes Yes Yes Region Dummies Yes Yes Yes Yes Observations 183 248 186 253 Pseudo R2 0.229 0.169 0.123 0.092 Logistics regressions; odds ratios reported; p-values in parentheses + p < 0.10, * p < 0.05, ** p < .01, *** p < .001 Programmatic parties and ICRG bureaucratic quality are measured at the project start year; all other variables are averaged over the project period. The base regressions pool all countries, more and less democratic, since our hypotheses apply equally to both. Table 4 shows that the results are robust, however, to restricting the sample to countries that have exhibited competitive elections and have had at least one democratic change of government; the definition follows, and the data are taken from, ACLP (Alvarez et. al. 1996, Cheibub et. al. 2009). Within democracies and controlling for IEG quality at entry rating and initial bureaucratic quality, a unit increase in the proportion of programmatic parties is associated with a nine-fold increase (8.68 ) in the odds of a successful institutional development impact rating relative to the odds of an unsuccessful rating, and a similar nine-fold increase (9.37) in the odds of a successful overall outcome rating relative to an unsuccessful rating. 34 Table 4: Programmatic Parties and Project Ratings, Democracy Subsample Base Model Full Model Base Model Full Model Inst. Dev. Impact Rating Overall Outcome Rating Programmatic 5.976* 8.688* 5.108** 9.367* Parties (0.020) (0.015) (0.007) (0.012) IEG Quality at 2.104*** 3.498** Entry (0.000) (0.001) ICRG Bureaucratic 0.717 0.407** Quality (0.177) (0.001) Baseline controls Yes Yes Yes Yes Region Dummies Yes Yes Yes Yes Observations 161 147 164 150 Pseudo R2 0.254 0.339 0.135 0.354 Logistics regressions, with the same specification/control variables as in Table 2. Odds ratios reported; p-values in parentheses + p < 0.10, * p < 0.05, ** p < .01, *** p < .001. GDP per capita and population are averaged over the project period; progression to secondary school is a country average; programmatic parties and ICRG bureaucratic quality are measured at the project start year Unobserved country characteristics and the association of programmatic parties and successful public sector reform The results in Tables 2-4 are the first robust evidence of an association between the organizational characteristics of political parties and the success of personnel and financial management reforms of a country’s public administration. However, while we can control for unobserved regional fixed effects that might influence both the presence of programmatic parties and the success of public sector reform, data coverage makes it less easy to control for fixed unobserved country characteristics that might also yield a spurious correlation between parties and reform. To identify the effects of parties based only on within-country variation in parties and reform success, we require many countries in the sample to exhibit multiple public sector reforms over a sufficiently long period. Of the 109 countries with public sector 35 reform projects in our sample, 92 have multiple projects. In principle, this should provide ample scope for identifying the effects of programmatic political parties using only within- country, over-time variation. However, our data exhibit relatively few countries with both multiple, non-overlapping projects and intra-country variation in the degree to which parties are programmatic. Of the 92 countries with multiple projects, only 17 exhibit no overlap among projects. The remainder exhibit overlap, indicating that multiple projects in those countries are undertaken within the same political context. 14 This leaves little within- country variation in the political circumstances under which projects are implemented with which to identify the effects of programmatic political parties. Here we present numerous arguments, including estimates that control for country fixed effects, that suggest that unobserved country characteristics spuriously drive our results. For example, reverse causality is implausible. As the earlier discussion makes clear, public sector reforms could precipitate the emergence of more organized and possibly programmatic parties. However, we measure party organization in the year that the public sector reform project was approved, well in advance of the start of the project and disbursement of funds. Successful reforms cannot be a contributing factor to whether parties are programmatic. In addition, the strong bureaucracies that Shefter or Tilly have in mind are largely absent in the sample. Indeed, this is one reason why the World Bank extends loans to them to encourage public sector reform. It is, therefore, implausible that strong bureaucracies prompted politicians to re-organize their political parties in these countries. 14 Of those countries exhibiting project overlap, the average overlap is substantial: on average, 22 percent of project-years overlap. 36 Measurement error might also be a spurious explanation of the association. Countries might receive good ratings on their public sector reform projects because of unobserved characteristics that also make them more likely to have programmatic political parties. World Bank evaluations could themselves be endogenous to the presence of loosely organized, clientelist political parties. 15 However, the controls in Table 2 make such sources of measurement bias an unlikely explanation for our results. For example, estimates control for the IEG rating of quality at entry, which should fully capture omitted country effects that might introduce evaluator bias. In addition, if unobserved country characteristics spuriously drove an association between our two dependent variables and the party variables, they should exert that same influence on other IEG ratings, including those that are unrelated to the domestic political conditions in the borrowing country. That is, programmatic parties would be expected to have a significant effect on all IEG ratings, even those on which, in theory, they should have no influence. One of these is the IEG rating of the quality of Bank supervision. Once a loan is signed, Bank staff supervise country progress towards meeting loan objectives. The second is the IEG rating of overall Bank performance, which takes into account both project design and supervision; the distribution of responses is nearly identical to that for supervision. Using these two IEG variables as the dependent variables in the base model of Table 2, however, indicates that programmatic parties are not significant determinants of IEG ratings of World Bank supervision and overall performance (results not reported in the interests of 15 This is unlikely. No project or other Bank document that we reviewed, including the IEG Report (2008), even mentioned political parties. 37 length). The programmatic party variable is not related to ratings with which it should, in fact, have no association. This reduces the concern that the relationship between programmatic parties and project success is driven by some unobserved factor or IEG bias that should influence all ratings. Table 5 indicates that the results are unlikely to be driven by unobserved country effects. The specifications are the same as those as in Table 2, with the addition of controls for country fixed effects. In three of the four specifications in Table 5, including both of the specifications for the more relevant institutional development impact rating, the programmatic party variable remains significant. Moreover, although the party variable is not significant in the base specification of the overall outcome rating (the third column in Table 5), the estimated odds ratio is almost identical to that estimated in Table 2. 38 Table 5: Programmatic Parties and Project Ratings for Public Sector Reform (Country Fixed Effects) Base Model Full Model Base Model Full Model Inst. Dev. Impact Rating Overall Outcome Rating Programmatic 7.269* 27.97** 2.608 23.02+ Parties (0.026) (0.004) (0.382) (0.063) Log GDP per 0.123 0.0350 0.446 0.163 capita (0.234) (0.133) (0.452) (0.502) Log Population 10.52 1414.6+ 0.0111 0.381 (0.486) (0.084) (0.158) (0.818) Log Land Area 0.785 6.987 284915.5 4.135 (0.916) (0.371) (0.122) (0.895) Funding Amount 0.994+ 0.993+ 0.997 0.994* (0.053) (0.052) (0.151) (0.023) Project Duration 0.964 1.077 0.801+ 0.922 (0.757) (0.608) (0.081) (0.687) Loan % Allocated 1.002 1.004 0.994 0.994 to PSR (0.872) (0.754) (0.574) (0.716) IEG Quality at 1.900* 9.324** Entry (0.015) (0.003) ICRG 0.483 0.510 Bureaucratic (0.327) (0.382) Quality Country FE Yes Yes Yes Yes Observations 148 130 196 168 Pseudo R 2 0.167 0.265 0.148 0.430 Odds ratios reported; p-values in parentheses + p < 0.10, * p < 0.05, ** p < .01, *** p < .001 GDP per capita and population are averaged over the project period; progression to secondary school is a country average; programmatic parties and ICRG bureaucratic quality are measured at the project start year Selection effects and the association of programmatic parties and successful public sector reform Selection bias would be a serious issue for the analysis here if we were asking whether countries are better able to reform their public sector if they receive a World Bank 39 public sector loan. For example, countries that ask for such loans could be more likely to reform than countries that do not. We would not be able to infer, from the experience of countries that accepted such loans, what would have happened to countries that did not apply for such loans, or rejected them. Our question is different, however: among those countries that apply for World Bank public sector loans, are public sector reforms more likely to succeed among those with programmatic parties than among those without? Selection bias affects the interpretation of our results only in the following unlikely scenario. Among those countries that apply for World Bank loans, those with programmatic parties are more likely than those with non- programmatic parties to successfully implement public sector reform. Among those that do not apply for loans, there is no difference between those with and without programmatic parties. In fact, our theory suggests that the bias should operate in the opposite direction. Countries with non-programmatic parties should be less likely to apply for public sector loans; the fact that they do indicates that they may have unobserved characteristics that make them more similar to the countries with programmatic parties that also apply for loans than to the non-programmatic countries that do not apply. To the extent that the countries with programmatic and non-programmatic countries that do apply for loans are more similar, however, we are less likely to observe the differences reported in Tables 2 – 4. Figure 3 provides further evidence that selection bias is an implausible explanation for our results. It shows that there are no significant differences across observable characteristics between countries that initiate public sector reforms and those that do not. The figure displays the subsample means and confidence intervals for the variables of 40 interest in this study. 16 These are our key independent variable, programmatic parties, logged GDP per capita and land area, as well as other institutional factors (competitiveness of legislative and executive elections, ACLP democracy coding, and whether the system is presidential or parliamentary). Scores for various governance indicators from International Country Risk Guide (bureaucratic quality, ethnic tension, corruption, and rule of law, all measured on the ICRG six-point scale) are also identical between the samples. In all cases, the means are within the overlapping areas of the respective confidence intervals, indicating that they are statistically indistinguishable between the two sets of countries. A further concern is that it is not party organization that drives the results we observe, but the different political processes that programmatic and non-programmatic parties may pursue prior to asking for World Bank loans. For example, it could be the case that countries that have programmatic parties agree to World Bank loans only after they have obtained broad support for them; those without such parties agree to the loans after a few top officials assent to them. Since the former process is more likely to yield successful loans, such a difference between programmatic and non-programmatic countries would yield a bias in favor of finding that programmatic parties lead to public sector reform success. In fact, though, all the countries in our sample share a similar legislative starting point. Legal authority to sign loan agreements with the World Bank rests with the executive branch and, within the executive, with the responsible sectoral minister and the finance minister. In the case of a health loan, for example, the minister of health and the minister of 16 For comparability, the subsample means are calculated for the range of country-years starting with the first year in which a country in the region engaged in reforms. Results are robust to using other ways of determining the time frame. For example, results are identical for using the first year that any country engaged in public sector reforms or for using the exact start date of reforms for the countries engaging in public sector reform. 41 finance are the usual Bank counter-parties. Very rarely do national legislatures, or even cabinets, approve World Bank loans. For essentially all of the sample countries and loans, therefore, loan agreements have proceeded after similar levels of support have been exhibited by officials occupying similar positions. Figure 3. Statistical balance between recipients of public sector reforms (PSR) and non-recipients Note: The figure illustrates the (standardized) means and 95 percent confidence intervals of 11 country characteristics. In all cases, the means are within the overlapping areas of the respective confidence intervals, indicating that they are statistically indistinguishable. The programmatic parties, executive and legislative elections and system variables are all from the Database of Political Institutions; log land area and per capita GDP are from World Development Indicators; democracy is from the ACLP database; Ethnic Tension, Rule of Law, Corruption and Bureaucratic Quality are all from International Country Risk Guide. 42 Discussion and implications for the literature on the political economy of public sector reform The analysis offers a new perspective on the political economy of politicians’ oversight of the bureaucracy. Prior research (e.g., McCubbins and Schwartz 1984, Moe 1987, Calvert, McCubbins and Weingast 1989, Thies 1989, Epstein and O’Halloran 1999) focuses on how legislators control administrative decisions of the executive branch, asking whether politicians rely on ex ante prescriptions (e.g., precise directions to the bureaucracy on what decisions to make under a range of specific circumstances) or ex post supervision (e.g., oversight committees and judicial review of administrative decisions). These control mechanisms share the common characteristic that they are systemic and assume that politicians can act collectively to enforce them. In the absence of programmatic political parties, however, none of these methods is credible; legislators have limited ability to act collectively to punish executives who ignore them. For example, the arguments here indicate considerably less interest in ex ante institutional rules among politicians who are not organized into programmatic parties. They are more likely to rely on personal contacts with affected constituents to find out whether adverse regulatory decisions have been made and to use their personal contacts with the bureaucracy, rather than formal legislative action, to monitor implementation and seek redress. Other research (e.g., Huber and McCarthy 2004) predicts that dysfunctional public administration, particularly agencies staffed with less capable employees, distorts politicians’ policy choices. The analysis here suggests that underlying political circumstances distort both political decisions regarding the bureaucracy and public policies more generally. Where politicians are not organized into programmatic political parties, they prefer clientelist public 43 policies (Keefer 2011) and rely on clientelist interactions with the public administration to ensure that these policies are carried out. The results here also extend the existing literature on the political economy of public administration to non-democracies. Autocrats vary in the degree to which they choose to fend off potential challengers by providing benefits to large groups of citizens. Those that choose to provide broadly distributed benefits to citizens must make credible commitments to reward members of the ruling party, who implement these policies. As Gehlbach and Keefer (2010) argue, the credibility of ruler commitments to party members depends on whether party members can act collectively to limit the scope for ruler shirking. However, party members able to act collectively are also more likely to demand such administrative reforms to facilitate oversight of the ruler’s pursuit of party goals. Public sector reforms are therefore also more likely to succeed in non-democracies with programmatic parties. Conclusion This paper develops a new argument that relates public sector organization to the organization of political parties: the better able are politicians to act collectively to prevent shirking by party leaders, the more likely are voters to hold them accountable for public sector failures and the greater are political incentives to enforce rules governing public sector performance. Such organizational arrangements are essential for the credibility of parties’ programmatic commitments. Using data on whether political parties are programmatic and new evidence on World Bank public sector loans, we find strong and robust evidence that public sector reforms are least likely to succeed in countries in which no trace of programmatic tendencies can be detected in their largest political parties. 44 These results are further evidence that the organization of political parties, particularly whether they are organized to allow leaders to prevent free-riding by members and members to act collectively to control shirking by leaders, has significant consequences for development phenomena. Our results also have significant implications for public sector reform. In discussing the timing of reform and of donor collaboration with client countries, practitioners and donors emphasize the importance of windows of opportunity. Those windows are usually considered open when influential politicians demonstrate enthusiasm for reform. The analysis here holds such enthusiasm constant: all loans generated enough enthusiasm to persuade a few top ministers, including the one in charge of the relevant sector, to approve the loan. The individual commitment of a few turns out to be insufficient, however. In addition, and critically, reform succeeds when a larger group of politicians benefits collectively from it –when politicians are gathered under the umbrella of political parties organized to promote their collective action. Data Appendix The IEG selected 467 loans that began in 1987 or later and ended in 2006 or earlier. World Bank loans often have multiple, cross-sectoral objectives. IEG therefore classified as “public sector” (as opposed to education, infrastructure, health, etc.) those projects that met two criteria. First, at least 25 percent of the loan amount had to be related to themes or sectors tied to public sector reform. Second, loans must be disbursed when at least three of the conditions for disbursement fall under the public sector rubric. For projects active since December 2001, the Bank’s management information system classifies projects according both to the sectors (public sector, economic policy, 45 education, etc.) and up to five sub-sector themes (e.g., tertiary education). 17 The most frequently assigned sub-sector themes in loans classified as public sector are public expenditure and financial management, appearing in 307 projects; administrative and civil service reform (249); state enterprise reform (144); and tax policy and administration (120). No others appear more than 90 times (e.g., regulation and competition policy, 90; debt management and fiscal sustainability, 87; and macroeconomic management, 83). Public sector reform in the World Bank includes themes (e.g., decentralization and judicial reform) that fit only loosely with the arguments here. In our base regressions we therefore restrict the analysis only to those loans for which at least one sub-sector theme was public expenditure and financial management; administrative and civil service reform; or tax policy and administration. Each of these maps directly into the earlier discussion. Public expenditure and financial management includes budget planning and execution, procurement, auditing, and monitoring and evaluation. For example, budget reforms might bring off-budget spending into the budget envelope, allowing for all politicians to monitor it. One World Bank document outlining the needs of Benin with respect to financial management, for example, emphasizes the need for integrated budget and accounts systems and improved budget management capacity in the Ministry of Finance and line ministries (World Bank 2001). Administrative and civil service reform involves all aspects of the management and organization of personnel, including personnel information systems, career paths, and pay systems. One World Bank-sponsored reform, in Armenia, for example, was an Electronic 17 Prior to December 2001, only sectoral classifications are available. 46 Dcoument Management System to allow the automation of work planning and the performance management process in the civil service (World Bank no date). Finally, tax administration is defined as including the key aspects of revenue administration, particularly the institutional setting and development of operational processes, including automation and interaction with taxpayers (World Bank 2008, 4-7). These particular public sector reform areas closely match the foregoing theoretical arguments. However, results reported below are robust to alternative samples, including the whole sample of loans (based on the argument that the World Bank classifies all of these as public sector loans); or after excluding tax policy and administration (since, although tax administration itself is entirely within the scope of our analysis, loans to assist in the revision of the tax code itself are less so). Having chosen the 467 (511 in the updated database used in this study) public sector reform loans, the IEG subjected them to several evaluations, as it does to all loans that it evaluates. Two are most appropriate for the analysis here. First, the IEG ratings evaluate institutional development impact. This is the extent to which a project improves the ability of a country or region to make more efficient, equitable and sustainable use of its human, financial, and natural resources. For loans in most sectors, the main objectives of the projects can be achieved with little attention to institutional development (dams can be built successfully even if the ministry in charge of them remains a shambles). In the case of public sector reforms, however, institutional development is precisely the point. Moreover, in loans with mixed objectives, some public sector and some not, the institutional development rating is likely to weight the public sector components most highly. 47 The second IEG rating used here is that of a loan’s overall outcomes, ranging from highly satisfactory (the program achieved at least acceptable progress toward all major relevant objectives and had best practice development impact on one or more of them); to satisfactory (progress on all relevant objectives, none best practice and no major shortcomings) to moderately satisfactory (major progress on most objectives and no major shortcomings) to moderately unsatisfactory (no progress on most major objectives or progress on all, but with a major shortcoming, such as a violation of a safeguard (e.g., such as the improper resettlement of communities in an area to be dammed); and unsatisfactory – did not make acceptable progress on most objectives and produced a major shortcoming. The “overall outcome” rating formed the basis of the IEG’s own evaluation of public sector reform loans. 48 References Acemoglu, Daron, Simon Johnson and James Robinson (2001). “The Colonial Origins of Comparative Development: An Empirical Investigation.” American Economic Review 91(5): 1369-1401. Aldrich, John H. (1995) Why Parties? The Origin and Transformation of Political Parties in America. Edited by B. I. Page, American Politics and Political Economy Series. Chicago: University of Chicago Press. Alvarez, Mike, José Antonio Cheibub, Fernando Limongi and Adam Przeworski (1996). "Classifying Political Regimes." Studies in Comparative International Development 31(Summer):3-36. Bawn, Kathleen (1997). “Choosing Strategies to Control the Bureaucracy: Statutory Constraints, Oversight and the Committee System.” Journal of Law, Economics and Organization 13: 101-126. Beck, Thorsten, George Clarke, Alberto Groff, Philip Keefer and Patrick Walsh (2001). “New Tools in Comparative Political Economy: The Database of Political Institutions." World Bank Economic Review, 15(1): 165-176. Bowler, Shaun, David M. Farrell and Richard S. Katz (1999). "Party Cohesion, Party Discipline, and Parliaments." Party Discipline and Parliamentary Government, Shaun Bowler, David M. Farrell and Richard S. Katz, eds., 3-22. Columbus: Ohio State University Press. Cain, Bruce John Ferejohn and Morris Fiorina (1987). The Personal Vote: Constituency Service and Electoral Independence. Cambridge: Harvard University Press. Calvert, Randall L., Mathew D. McCubbins and Barry R. Weingast (1989). “A Theory of 49 Political Control and Agency Discretion.” American Journal of Political Science 33:3, 588-611. Calvo, Ernesto and M. Victoria Murillo (2013, forthcoming). “When Parties Meet Voters: Partisan Networks and Distributive Expectations in Argentina and Chile.” Comparative Political Studies. Chandra, Kanchan (2004). Why Ethnic Parties Succeed. Cambridge: Cambridge University Press. Cheibub, Jose Antonio, Jennifer Gandhi, and James Raymond Vreeland. 2009. Democracy and Dictatorship Revisited. Public Choice 143(1-2): 67-101. Cox, Gary W (1997). Making Votes Count: Strategic Coordination in the World’s Electoral Systems. Cambridge: Cambridge University Press, 1997. Cox, Gary W. and Mathew D. McCubbins (1994). “Bonding, Structure, and the Stability of Political Parties: Party Government in the House.” Legislative Studies Quarterly 19:2, 215-231. _____ (1993). Legislative Leviathan: Party Government in the House. Berkeley: University of California Press. Cragg, John G. and Stephen G. Donald (1993). “Testing Identifiability and Specification in Instrumental Variable Models,” Econometric Theory 9, 222 – 240. Epstein, David and Sharyn O’Halloran (1999). Delegating Powers: A Transaction-Cost Approach to Policymaking under Separate Powers. Cambridge: Cambridge University Press. Ferejohn, John and Charles Shipan (1990). “Congressional Influence on the Bureaucracy.” Journal of Law, Economics and Organization 6:1-20. Gehlbach, Scott and Philip Keefer (2011). “Investment without Democracy: Ruling-Party 50 Institutionalization and Credible Commitment in Autocracies.” Journal of Comparative Economics 39:2, 123-278 (June). Huber, Gregg and Nolan McCarty (2004). “Bureaucratic Capacity, Delegation, and Political Reform.” American Political Science Review 98: 481-94. Ishii, Risako, Piyawadee Rohitarachoon and Farhad Hossain (no date). “HRM [Human Resource Management] Reform in Decentralized Local Government: Empirical Perspectives on Recruitment and Selection in the Philippines and Thailand.” Mimeo, The University of Tokyo. Johnson, Ronald and Gary D. Libecap (1994). The Federal Civil Service System and the Problem of Bureaucracy. Chicago: University of Chicago Press. Keefer, Philip (2011). “Collective Action, Political Parties and Pro-Development Public Policy.” Asian Development Review 28:1, 94-118. _____ (Forthcoming). “Organizing for prosperity: Collective action, political parties and the political economy of development” Oxford Handbook of the Politics of Development, Carol Lancaster and Nicholas van de Walle, editors. _____ and Razvan Vlaicu (2008). “Democracy, Credibility and Clientelism.” Journal of Law, Economics and Organization 24:2, 371-406 (October). _____ and Stuti Khemani (2009). “When Do Legislators Pass on Pork? The Role of Political Parties in Determining Legislator Effort.” American Political Science Review 103:199-112 (February). Kitschelt, Herbert (1993). “Social Movements, Political Parties, and Democratic Theory.” Annals of the American Academy of Political and Social Sciences. 528:13 - 29. _____ (2000). “Linkages between citizens and politicians in democratic polities.” 51 Comparative Political Studies. 33:6-7, 845-879. Knack, Steven (2008). “Sovereign Rents and the Quality of Tax Policy and Administration.” World Bank Policy Research Working Paper 4773 (November). _____ and Philip Keefer (1995). "Institutions and Economic Performance: Cross-Country Tests Using Alternative Institutional Measures." Economics and Politics 7(3): 207-228. McCubbins, Mathew and Thomas Schwartz (1984). “Congressional Oversight Overlooked: Police Patrols versus Fire Alarms.” American Journal of Political Science 28: 165-79 McCubbins, Mathew, Roger Noll and Barry Weingast (1987). “Administrative Procedures as Instruments of Political Control.” Journal of Law, Economics and Organization 3:243 – 77. McCubbins, Mathew and Frances Rosenbluth (1995). "Party Provision of Personal Politics: Dividing the Vote in Japan." Structure and Policy in Japan and the U.S., Peter Cowhey and Mathew McCubbins, eds., 35-55. Cambridge: Cambridge University Press. Moe, Terry M. (1987). “An Assessment of the Positive Theory of ‘Congressional Dominance.’” Legislative Studies Quarterly 12:4, 475-520. Rauch, James and Peter Evans (2000). “Bureaucratic structures and economic performance in less developed countries.” Journal of Public Economics 75, 49–71. Shefter, Martin (1994). Political Parties and the State: The American Historical Experience. Princeton, New Jersey: Princeton University Press. Snyder Jr., James M., and Michael M. Ting (2002). “An Informational Rationale for Political Parties.” American Journal of Political Science 46 (1):90-110. Stock, James and Motohiro Yogo (2004). “Testing for Weak Instruments in Linear IV Regressions.” Festschrift in Honor of Thomas Rothenberg, D.W.K. Andrews and James 52 Stock, eds. Cambridge: Cambridge University Press. Stokes, Susan C. (2003). “Perverse Accountability: A Formal Model of Machine Politics with Evidence from Argentina.” American Political Science Review 99:3, 315-325 (August). Thies, Michael F. (2001). “Keeping Tabs on Partners: The Logic of Delegation in Coalition Governments.” American Journal of Political Science 45:3, 580-98. Tilly, Charles (1990). Coercion, Capital and European States, AD 990-1990. Cambridge: Basil Blackwell. van de Walle, Nicolas (2003). “Presidentialism and Clientelism in Africa's Emerging Party Systems.” Journal of Modern African Studies 41 (2):297-321. Weingast, Barry R., and Moran, Mark J. (1983). "Bureaucratic Discretion or Congressional Control? Regulatory Policymaking by the Federal Trade Commission." Journal of Political Economy 91 (October): 765-800. World Bank (2001). “Republic of Benin: Country financial accountability assessment.” Report 26772. June 27. World Bank (2008). Public Sector Reform: What Works and Why? An IEG Evaluation of World Bank Support. Washington DC: International Bank for Reconstruction and Development/The World Bank. _____ (no date). “Governance and Public Sector Management.” http://siteresources.worldbank.org/PUBLICSECTORANDGOVERNANCE/Res ources/285741-1348676611027/PSGDataSheet.pdf 53